from fastapi import FastAPI, File, UploadFile, HTTPException, Request from fastapi.middleware.cors import CORSMiddleware from fastapi.responses import JSONResponse import tensorflow as tf import numpy as np import cv2 from PIL import Image import os import warnings import base64 import io from pydantic import BaseModel from typing import List, Optional, Dict warnings.filterwarnings('ignore') # Initialize FastAPI app app = FastAPI(title="AI Fashion Recommendation API") # Configure CORS app.add_middleware( CORSMiddleware, allow_origins=["*"], allow_credentials=True, allow_methods=["*"], allow_headers=["*"], ) # Configure TensorFlow to use CPU only tf.config.set_visible_devices([], 'GPU') os.environ['CUDA_VISIBLE_DEVICES'] = '-1' # Define face shape labels face_shape_labels = ['Heart', 'Oblong', 'Oval', 'Round', 'Square'] # Global variables for models face_detection_model = None # Define the model path (update this path according to your setup) model_path = './Try_Face_Detection_AI_1.keras' # Update this path # Pydantic models for request validation class TextRecommendationRequest(BaseModel): gender: str skin_tone: str age_group: str categories: List[str] class Base64ImageRequest(BaseModel): image_base64: str categories: List[str] ############################################################## # FACE DETECTION AND PROCESSING FUNCTIONS ############################################################## def detect_face_with_opencv(image): """Detect face using OpenCV's Haar Cascade""" if image is None: return None if not isinstance(image, np.ndarray): if hasattr(image, 'convert'): image = np.array(image.convert('RGB')) else: image = np.array(image) gray = cv2.cvtColor(image, cv2.COLOR_RGB2GRAY) face_cascade_path = cv2.data.haarcascades + 'haarcascade_frontalface_default.xml' if not os.path.exists(face_cascade_path): raise HTTPException(status_code=500, detail="Haar cascade file not found") face_cascade = cv2.CascadeClassifier(face_cascade_path) faces = face_cascade.detectMultiScale(gray, scaleFactor=1.1, minNeighbors=5, minSize=(30, 30)) if len(faces) > 0: x, y, w, h = faces[0] return image[y:y+h, x:x+w] return None def extract_face(image): face_img = detect_face_with_opencv(image) if face_img is not None: return cv2.resize(face_img, (224, 224)) print("WARNING: Could not detect face with OpenCV") if isinstance(image, np.ndarray): return cv2.resize(image, (224, 224)) elif hasattr(image, 'resize'): return np.array(image.resize((224, 224))) return None def preprocess_image(image): try: if isinstance(image, np.ndarray): rgb_image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB) if len(image.shape) == 3 and image.shape[2] == 3 else image else: rgb_image = np.array(image.convert('RGB')) if hasattr(image, 'convert') else np.array(image) if rgb_image.shape[0] != 224 or rgb_image.shape[1] != 224: rgb_image = cv2.resize(rgb_image, (224, 224)) if len(rgb_image.shape) == 2: rgb_image = cv2.cvtColor(rgb_image, cv2.COLOR_GRAY2RGB) elif rgb_image.shape[2] == 4: rgb_image = cv2.cvtColor(rgb_image, cv2.COLOR_RGBA2RGB) normalized_image = rgb_image / 255.0 return np.expand_dims(normalized_image, axis=0) except Exception as e: raise HTTPException(status_code=500, detail=f"Image preprocessing failed: {str(e)}") def load_face_shape_model(): global face_detection_model try: with tf.device('/CPU:0'): face_detection_model = tf.keras.models.load_model(model_path) print("Model loaded successfully!") except Exception as e: print(f"Warning: Could not load model: {e}") face_detection_model = tf.keras.Sequential([ tf.keras.layers.Input(shape=(224, 224, 3)), tf.keras.layers.Conv2D(16, 3, activation='relu'), tf.keras.layers.GlobalAveragePooling2D(), tf.keras.layers.Dense(5, activation='softmax') ]) print("Created dummy model") def predict_face_shape(image): global face_detection_model if image is None: raise HTTPException(status_code=400, detail="No image provided") face_image = extract_face(image) if face_image is None: return {"face_shape": "Oval", "confidence": 50.0, "note": "Default due to face detection error"} try: preprocessed_image = preprocess_image(face_image) with tf.device('/CPU:0'): predictions = face_detection_model.predict(preprocessed_image) predicted_class = np.argmax(predictions) confidence = float(predictions[0][predicted_class]) * 100 return { "face_shape": face_shape_labels[predicted_class], "confidence": round(confidence, 1) } except Exception as e: print(f"Prediction error: {e}") return {"face_shape": "Oval", "confidence": 50.0, "note": "Default due to prediction error"} ############################################################## # RECOMMENDATION DATA (Same as original) ############################################################## face_shape_recommendations = { "Heart": { "Glasses": [ "Cat Eye Frames", "Round Frames", "Clear Frames", "Oval Glasses", "Alford Glasses", "Tortoiseshell Sunglasses", "Transparent Eyeglasses Frames", "Geometric Frames", "Aviator Glasses", "Clubmaster Frames", "Oversized Glasses", "Square Frames", "Wayfarer Glasses", "Browline Glasses", "Rimless Glasses", "Classic Aviators", "Butterfly Frames", "Pantos Frames", "Pilot Glasses", "Rectangle Frames" ], "Watches": [ "Luxury Watch", "Minimalist Watch", "Chronograph Watch", "Pilot Watch", "Diver Watch", "Sveston Sports Watch", "Casio G-Shock", "Casio Edifice", "Casio Protrek", "Fossil Silicon Watch", "Swiss Military Alpine", "Hanowa Puma Watch", "Swiss Chronograph", "Smart BT Calling Watch", "Infinity Smart Watch", "Vogue Smart Watch", "Realme Watch S2", "Mibro Watch C4", "Redmi Watch 5", "Bold Dial Watch" ], "Hats": [ "Beanie", "Wide-Brim Hat", "Trilby", "Newsboy Cap", "Cowboy Hat", "Trucker Hat", "Safari Hat", "Flat Cap", "Boater Hat", "Top Hat", "Classic Fedora", "Chitrali Cap", "Gilgiti Cap", "Pakol", "Baseball Cap", "Snapback Cap", "Bucket Hat", "Beret", "Panama Hat", "Pork Pie Hat" ] }, "Oblong": { "Glasses": [ "Aviators", "Oversized Glasses", "Round Frames", "Square Frames", "Wayfarer Glasses", "Tortoiseshell Sunglasses", "Transparent Eyeglasses Frames", "Geometric Frames", "Cat Eye Frames", "Clubmaster Frames", "Oval Glasses", "Clear Frames", "Butterfly Frames", "Pantos Frames", "Pilot Glasses", "Rectangle Frames", "Browline Glasses", "Rimless Glasses", "Classic Aviators", "Embellished Sunglasses" ], "Watches": [ "Pilot Watch", "Luxury Watch", "Minimalist Watch", "Chronograph Watch", "Diver Watch", "Sveston Sports Watch", "Casio G-Shock", "Casio Edifice", "Casio Protrek", "Fossil Silicon Watch", "Swiss Military Alpine", "Hanowa Puma Watch", "Swiss Chronograph", "Smart BT Calling Watch", "Infinity Smart Watch", "Vogue Smart Watch", "Realme Watch S2", "Mibro Watch C4", "Redmi Watch 5", "Bold Dial Watch" ], "Hats": [ "Trilby", "Newsboy Cap", "Cowboy Hat", "Safari Hat", "Flat Cap", "Trucker Hat", "Beanie", "Wide-Brim Hat", "Boater Hat", "Top Hat", "Classic Fedora", "Chitrali Cap", "Gilgiti Cap", "Pakol", "Baseball Cap", "Snapback Cap", "Bucket Hat", "Beret", "Panama Hat", "Pork Pie Hat" ] }, "Oval": { "Glasses": [ "Wayfarer Glasses", "Geometric Frames", "Cat Eye Frames", "Round Frames", "Clear Frames", "Aviator Glasses", "Clubmaster Frames", "Square Frames", "Oversized Glasses", "Oval Glasses", "Transparent Frames", "Tortoiseshell Frames", "Browline Glasses", "Classic Aviators", "Butterfly Frames", "Rimless Glasses", "Rectangle Frames", "Pilot Glasses", "Metal Frame Glasses", "Gradient Sunglasses" ], "Watches": [ "Diver Watch", "Dress Watch", "Luxury Watch", "Minimalist Watch", "Chronograph Watch", "Smart BT Calling Watch", "Realme Watch S2", "Fossil Gen 6 Smartwatch", "Casio Edifice", "Swiss Military Alpine", "Sveston Classic", "Hanowa Chronograph", "Infinity Smart Watch", "Mibro T1 Smartwatch", "Vogue Smart Watch", "T500+ Smart Watch", "Casio F91W", "Xiaomi Watch 2", "Skeleton Watch", "Bold Dial Watch" ], "Hats": [ "Cowboy Hat", "Safari Hat", "Trilby", "Newsboy Cap", "Flat Cap", "Wide-Brim Hat", "Boater Hat", "Top Hat", "Classic Fedora", "Pakol", "Gilgiti Cap", "Baseball Cap", "Bucket Hat", "Snapback Cap", "Beret", "Panama Hat", "Pork Pie Hat", "Sun Hat", "Chitrali Cap", "Trucker Hat" ] }, "Round": { "Glasses": [ "Square Frames", "Browline Glasses", "Cat Eye Frames", "Round Frames", "Clear Frames", "Wayfarer Glasses", "Geometric Frames", "Clubmaster Frames", "Rectangle Frames", "Tortoiseshell Frames", "Metal Frame Glasses", "Oversized Glasses", "Aviator Glasses", "Butterfly Frames", "Classic Aviators", "Transparent Frames", "Rimless Glasses", "Oval Glasses", "Pilot Glasses", "Gradient Sunglasses" ], "Watches": [ "Bold Dial Watch", "Square Dial Watch", "Luxury Watch", "Minimalist Watch", "Chronograph Watch", "Casio G-Shock", "Sveston Classic Watch", "Swiss Military Alpine", "Hanowa Smart Watch", "Infinity Smart Watch", "Fossil Smart Watch", "Realme Watch S2", "Mibro T1 Smartwatch", "Dress Watch", "Smart BT Calling Watch", "Casio Edifice", "Vogue Smart Watch", "T500+ Smart Watch", "Skeleton Watch", "Retro Watch" ], "Hats": [ "Flat Cap", "Boater Hat", "Trilby", "Newsboy Cap", "Cowboy Hat", "Wide-Brim Hat", "Safari Hat", "Classic Fedora", "Pakol", "Chitrali Cap", "Snapback Cap", "Bucket Hat", "Top Hat", "Baseball Cap", "Panama Hat", "Pork Pie Hat", "Sun Hat", "Beret", "Trucker Hat", "Gilgiti Cap" ] }, "Square": { "Glasses": [ "Rimless Glasses", "Classic Aviators", "Cat Eye Frames", "Round Frames", "Clear Frames", "Wayfarer Glasses", "Geometric Frames", "Clubmaster Frames", "Square Frames", "Tortoiseshell Glasses", "Aviator Glasses", "Browline Glasses", "Transparent Frames", "Butterfly Frames", "Rectangle Frames", "Pilot Glasses", "Metal Frame Glasses", "Oversized Frames", "Oval Glasses", "Gradient Sunglasses" ], "Watches": [ "Skeleton Watch", "Retro Watch", "Luxury Watch", "Minimalist Watch", "Chronograph Watch", "Dress Watch", "Casio Edifice", "Smart BT Calling Watch", "Infinity Smart Watch", "Realme Watch S2", "Fossil Gen 6", "Mibro T1", "Swiss Military Alpine", "Hanowa Puma Watch", "Casio G-Shock", "Redmi Watch 5", "Vogue Smart Watch", "Bold Dial Watch", "Square Dial Watch", "Pilot Watch" ], "Hats": [ "Top Hat", "Classic Fedora", "Trilby", "Newsboy Cap", "Cowboy Hat", "Flat Cap", "Safari Hat", "Boater Hat", "Snapback Cap", "Bucket Hat", "Baseball Cap", "Panama Hat", "Pork Pie Hat", "Beret", "Sun Hat", "Wide-Brim Hat", "Trucker Hat", "Chitrali Cap", "Pakol", "Gilgiti Cap" ] } } ############################################################## # API ENDPOINTS ############################################################## @app.on_event("startup") async def load_model(): print("Loading face shape detection model...") load_face_shape_model() print("API ready!") @app.get("/", tags=["Root"]) async def home(): return { "message": "AI Fashion Recommendation API is running!", "version": "1.0", "endpoints": { "image_recommendations": "/predict/image", "text_recommendations": "/predict/text", "face_shape_detection": "/detect/face-shape" } } @app.post("/predict/image", tags=["Predictions"]) async def image_recommendations( request: Request, image: Optional[UploadFile] = File(None), categories: List[str] = [] ): try: # Handle base64 or file upload if await request.body(): data = await request.json() if 'image_base64' in data: image_data = base64.b64decode(data['image_base64']) image = Image.open(io.BytesIO(image_data)) categories = data.get('categories', []) else: raise HTTPException(status_code=400, detail="No image provided") elif image: contents = await image.read() image = Image.open(io.BytesIO(contents)) else: raise HTTPException(status_code=400, detail="No image provided") if not categories: raise HTTPException(status_code=400, detail="Select at least one category") face_shape_result = predict_face_shape(image) face_shape = face_shape_result.get("face_shape", "Oval") recommendations = {} for category in categories: recs = face_shape_recommendations.get(face_shape, {}).get(category, []) recommendations[category] = recs[:5] return { "face_shape_info": face_shape_result, "recommendations": recommendations, "categories": categories } except Exception as e: raise HTTPException(status_code=500, detail=str(e)) @app.post("/predict/text", tags=["Predictions"]) async def text_recommendations(request: TextRecommendationRequest): try: recommendations = {} for category in request.categories: recs = face_shape_recommendations.get("Oval", {}).get(category, []) recommendations[category] = recs[:5] return { "user_attributes": request.dict(), "recommendations": recommendations, "note": "General fashion trends" } except Exception as e: raise HTTPException(status_code=500, detail=str(e)) @app.post("/detect/face-shape", tags=["Detection"]) async def detect_face_shape( image: Optional[UploadFile] = File(None), request: Optional[Base64ImageRequest] = None ): try: if request and request.image_base64: image_data = base64.b64decode(request.image_base64) image = Image.open(io.BytesIO(image_data)) elif image: contents = await image.read() image = Image.open(io.BytesIO(contents)) else: raise HTTPException(status_code=400, detail="No image provided") return predict_face_shape(image) except Exception as e: raise HTTPException(status_code=500, detail=str(e)) @app.get("/categories", tags=["Metadata"]) async def get_categories(): return { "categories": ["Glasses", "Watches", "Hats"], "face_shapes": face_shape_labels, "gender_options": ["Male", "Female", "Kid", "Transgender"], "skin_tone_options": ["Fair", "Medium", "Dark"], "age_group_options": ["Child (0-12)", "Teen (13-19)", "Young Adult (20-35)", "Adult (36-50)", "Senior (51+)"] } if __name__ == "__main__": import uvicorn uvicorn.run(app, host="0.0.0.0", port=5000)